Structure of the mitochondrial ATP synthase fromPichia angusta determined by electron cryo-microscopyKutti R. Vinothkumara, Martin G. Montgomeryb, Sidong Liub, and John E. Walkerb,1

aThe Medical Research Council Laboratory of Molecular Biology, Cambridge Biomedical Campus, Cambridge CB2 0QH, United Kingdom; and bThe MedicalResearch Council Mitochondrial Biology Unit, Cambridge Biomedical Campus, Cambridge CB2 0XY, United Kingdom

Contributed by John E. Walker, September 27, 2016 (sent for review July 18, 2016; reviewed by Wah Chiu and Stanley D. Dunn)

The structure of the intact monomeric ATP synthase from thefungus, Pichia angusta, has been solved by electron cryo-micros-copy. The structure provides insights into the mechanical couplingof the transmembrane proton motive force across mitochondrialmembranes in the synthesis of ATP. This mechanism requiresa strong and integral stator, consisting of the catalytic α3β3-domain, peripheral stalk, and, in the membrane domain, subunit aand associated supernumerary subunits, kept in contact with therotor turning at speeds up to 350 Hz. The stator’s integrity is en-sured by robust attachment of both the oligomycin sensitivity con-ferral protein (OSCP) to the catalytic domain and the membranedomain of subunit b to subunit a. The ATP8 subunit provides anadditional brace between the peripheral stalk and subunit a. Atthe junction between the OSCP and the apparently stiff, elongatedα-helical b-subunit and associated d- and h-subunits, an elbow orjoint allows the stator to bend to accommodate lateral movementsduring the activity of the catalytic domain. The stator may alsoapply lateral force to help keep the static a-subunit and rotatingc10-ring together. The interface between the c10-ring and the a-sub-unit contains the transmembrane pathway for protons, and theirpassage across the membrane generates the turning of the rotor.The pathway has two half-channels containing conserved polar res-idues provided by a bundle of four α-helices inclined at ∼30° to theplane of the membrane, similar to those described in other species.The structure provides more insights into the workings of thisamazing machine.

Pichia angusta | ATP synthase | structure | proton translocation

The ATP synthases (F-ATPases) found in mitochondria,chloroplasts, and eubacteria are membrane-bound molecular

machines with a rotary action. Our understanding of how theywork has come mainly from single-molecule studies of rotation onbacterial F-ATPases (1) and from structures of their constituentdomains determined predominantly with enzymes from mito-chondria (2–5). The most complete atomic resolution structurecontains about 85% of the bovine F-ATPase, including themembrane domain of the rotor. It is a mosaic built from sub-structures determined by X-ray crystallography (2) within theconstraints of an overall structure determined by cryo-electronmicroscopy (cryo-EM) (6). In this mosaic structure, the enzyme’srotor is defined clearly, but a detailed description of the remaining15% in the membrane domain of the stator and details of itsmembrane extrinsic domain are still needed as the basis for acomplete molecular understanding of the enzyme’s mechanism. Ageneral outline of some of the features of the residual region hasbeen provided by cryo-EM structures of the bovine (7) and analgal enzyme (8) and by the crystal structure of an intact bacterialenzyme (9), but none of them is at a sufficient resolution toprovide a complete molecular description of the enzyme.The mosaic structure (2) established that the rotor is an ensemble

of a membrane intrinsic domain, made of a ring of c-subunits, at-tached to a globular, elongated, and asymmetric structure, known asthe central stalk, made in mitochondria of single copies of subunitsγ, δ, and e. Its elongated region is a coiled-coil of α-helices in theN- and C-terminal regions of the γ-subunit, which penetrates into

the globular α3β3-catalytic domain. The α3β3-domain is part of thestator, against which the rotor turns, and is linked by the peripheralstalk to the a-subunit and six associated supernumerary subunitsin the membrane domain. The supernumerary subunits have noknown direct role in ATP synthesis, but some of them mediateinteractions between monomeric F-ATPase complexes in dimers(10–13) of the complex that are associated in rows along the edgesof the mitochondrial cristae (14, 15). The peripheral stalk itself ismade of single copies of the oligomycin sensitivity conferral protein(OSCP), subunit F6 (or the orthologous fungal h-subunit), and theb- and d-subunits (16–18). In the mosaic model (2), and reiteratedin the bovine cryo-EM structure (7), the OSCP is bound to theN-terminal region of one of the α-subunits, and the peripheral stalkextends via a largely α-helical structure in the b-, d-, and F6-subunitsalong the external surface of the F1-domain into the membraneregion of the enzyme. Here, the membrane sector of the b-subunit isthought to be bound to the a-subunit, keeping the static a-subunitand the rotating c-ring in contact and maintaining a specific path-way in their interface region for translocation of protons throughthe membrane. In the course of proton translocation, potentialenergy stored by the proton-motive force and the membrane ca-pacitance is released, impelling the turning of the rotor at speeds ofup to 350 cycles/s (19). During ATP synthesis, the turning of therotor in an anticlockwise direction as viewed from above the com-plex modulates the conformation of the three catalytic sites in theα3β3-domain, taking each of them through a cycle of substratebinding, ATP formation, and product release. Thus, each 360° ro-tation produces three molecules of ATP.

Significance

Living cells need fuel in the form of adenosine triphosphate, orATP, to stay alive. This fuel is generated by a molecular machinemade of two motors joined by a rotor. One generates rotation byusing energy provided by oxidative metabolism or photosynthe-sis; the other uses energy transmitted by the rotor to make ATPmolecules from its building blocks, adenosine diphosphate, orADP, and inorganic phosphate. The structure has been determinedof a fungal machine, isolated from its cellular power stations, themitochondria, where the machine operates. It provides un-suspected details of the blueprint of the machine and how itworks. The working principles of the fungal machine apply tosimilar machines in all species.

Author contributions: J.E.W. designed research; K.R.V., M.G.M., and S.L. performed re-search; K.R.V., M.G.M., and J.E.W. analyzed data; and K.R.V., M.G.M., and J.E.W. wrotethe paper.

Reviewers: W.C., Baylor College of Medicine; and S.D.D., University of Western Ontario.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.

Data deposition: The cryo-EM maps have been deposited in the EMDataBank (accessionnos. EMD-4102, EMD-4101, and EMD-4100). The atomic coordinates have been depositedin the Protein Data Bank (accession nos. 5LQZ, 5LQY, and 5LQX).1To whom correspondence should be addressed. Email: walker@mrc-mbu.cam.ac.uk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1615902113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1615902113 PNAS | November 8, 2016 | vol. 113 | no. 45 | 12709–12714

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 3,

202

1

As described here by cryo-EM, we have determined the struc-ture of the entire monomeric F-ATPase complex from the mito-chondria of the moderately thermophilic fungus, Pichia angusta,bound to the inhibitory region of the natural bovine inhibitorprotein, IF1 (inhibitor of F1-ATPase). This structure fills signifi-cant gaps in our knowledge of the mechanism of the F-ATPase.First, it contributes to our understanding of the coupling mecha-nism by providing unsuspected details of the peripheral stalk andhow it is attached to both the catalytic F1-domain and the mem-brane sector of the enzyme. Second, it provides an independentdescription of the transmembrane proton pathway for protons,helping both to define common features in the proton trans-location mechanism in bacterial and mitochondrial F-ATPasesand to explain human pathological mutations in the subunit.

ResultsStructure Determination. The structure was determined with themonomeric enzyme from P. angusta inhibited with residues 1–60of bovine IF1 (20). Formation of the inhibited complex requiresthe hydrolysis of ATP, and both bovine and yeast IF1 bind to theirorthologous F-ATPases at the same major site (3, 20). In struc-tures of F1-IF1 complexes, the inhibitor protein is entrapped in theαDPβDP-catalytic interface, the most closed of the three catalyticinterfaces of the enzyme (3–5, 21). In the intact F-ATPase, thissite can be in any one of three positions relative to the peripheralstalk, separated from each other by 120° rotations about thecentral axis of the F1-domain. The purified inhibited complexcontains all of the known subunits of the enzyme, excepting su-pernumerary subunits e, g, k, and l (20). Thus, the membranedomain of this preparation contains the c-ring and subunits a, f, j,ATP8, and the membrane domain of subunit b.The inhibited F-ATPase complex was examined by single-particle

cryo-EM (SI Appendix, Fig. S1). After refinement, motion correc-tion, and B-factor weighting, the entire dataset of 100,724 particlesresulted in a map with overall resolution of ∼7 Å. Following 3Dclassification of the weighted particles with the resolution of align-ment limited to 12 Å, the particles were grouped into three statesbased on the position of the inhibitor protein (SI Appendix, Figs. S2and S3). In state 1, the αDPβDP-catalytic interface with the bound

inhibitor, is closest to the peripheral stalk (SI Appendix, Fig. S2). Instates 2 and 3, the same interface and bound inhibitor are in the twoother possible orientations relative to the peripheral stalk. As 44.5%of the particles are in state 1, the resolution of the correspondingmap is superior to those of states 2 and 3, containing 22.8 and17.4% of the particles, respectively (SI Appendix, Fig. S2). The localresolution of the F1 domain is better than that of the membranedomain (SI Appendix, Fig. S4), and the detergent-lipid annulusaround the membrane domain has the lowest resolution. After re-finement, as expected, the three states differed mainly in the centralstalk and the F1-domain (SI Appendix, Fig. S2).The map from state 1 was used to build a model of the

F-ATPase, and states 2 and 3 were modeled from state 1. Thestate 1 map was interpreted with the structures of the F1-IF1complex (3) and the c-ring from the F1-c10 subcomplex (22), bothfrom Saccharomyces cerevisiae, and with the bovine peripheralstalk (18) and IF1 (21); as expected, α-helices are resolved moreclearly than β-strands. The final model (Fig. 1; SI Appendix, Fig. S5)contains the following residues: chains αE, 5–509; αTP, 12–406 and412–509; αDP, 7–509; βDP, 6–475; βE, 8–475; βTP, 7–475; γ, 1–59 and71–276; δ, 11–23 and 27–137; e, 1–49 and 53–61; bovine IF1 8–50;the c-subunits in the c10-ring, 3–75, 2–72, 1–72, 1–72; 1–73, 1–74,1–74, 1–73, 1–73, 2–73; OSCP, 6–153, and 180–194; b-subunit49–208; d-subunit 11–127, plus a segment of uncertain registermodeled as poly-Ala (1,001–1,028); h-subunit, modeled as poly-Ala(1,001–1,021); a-subunit, 44 residues in α-helices aH2 and aH3 (allmodeled as poly-Ala; 1,001–1,044); aH4 and aH5, 119–206; andaH6, 210–252. The model of the membrane domain also containsfour unconnected segments of secondary structure that have beenassigned, but with less certainty (see proposed identities in Fig. 1).They are the following: chain 1, 30 residues (1,001–1,030); chain2, 25 residues (1,001–1,025); chain 3, 17 residues (1,001–1,017);and chain 4, 27 residues (1,001–1,027), all modeled as poly-Ala. Allresidues have been truncated to Cβ throughout. At the secondarystructure level, there is excellent fit between the map and model(SI Appendix, Fig. S4), and the state 1 model was used in the in-terpretations below. The quality of the map is illustrated by bovineIF1 where residues 8–50 are well resolved (SI Appendix, Fig. S6). Inthe bovine cryo-EM structure (7), IF1 was not modeled.

BA C

Fig. 1. The structure of the F-ATPase from P. angusta. The enzyme was inhibited with residues 1–60 of the bovine inhibitor protein IF1. (A) The cryo-EM map ofstate 1 and structural model viewed from the side, with the peripheral stalk on the left, the catalytic domain at the top, attached by the central and peripheralstalks to the membrane domain below. (B and C) Side views of the enzyme-inhibitor complex in cartoon and surface representation, respectively. C is rotated tothe right by 90° relative to A and B. The α-, β-, γ-, δ-, and e-subunits forming the membrane extrinsic catalytic domain are red, yellow, royal blue, green, andmagenta, respectively; the inhibitor protein is cyan; and the peripheral stalk subunits OSCP, b, d, and h are sea-green, pink, orange, and purple, respectively. In themembrane domain, the c10-rotor is gray, the resolved region of the associated subunit a is corn-flower blue. Chains Ch1–Ch4 are pale yellow, brick-red, pale cyan,and beige, respectively, and have been assigned as transmembrane α-helices in subunit f and, in ATP8, as aH1 and bH1, respectively. In SI Appendix, Fig. S5, theidentities of subunits are placed directly on an enlarged version of C.

12710 | www.pnas.org/cgi/doi/10.1073/pnas.1615902113 Vinothkumar et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 3,

202

1

The Peripheral Stalk. In the structure of bovine F1-ATPase with atruncated version of the peripheral stalk (referred to as bF1-tPS)(18), containing the OSCP and F6 subunits, and residues 99–214and 1–118 of subunits b and d, respectively, the peripheral stalkis attached to the “top” of the F1-domain by the N-terminalregion of the αE-subunit interacting with the N-terminal domainof the OSCP, and the equivalent region in the EM map of thebovine enzyme has been modeled similarly (7). Residues 5–17and 7–20, respectively, of α-subunits in the bovine and P. angustaenzymes are predicted to be α-helical (SI Appendix, Fig. S7), butin models of bovine F1-ATPase alone, all three are largely un-structured (23–25). As in the bovine protein, the N-terminaldomain of the OSCP in P. angusta consists of a bundle of sixα-helices (OH1–OH6), and the N-terminal regions of two of thethree α-subunits interact with it. Like bF1-tPS, an α-helix con-taining residues 12–21 of the αE-subunit occupies a groove be-tween OH1 and OH5. In addition, a region of density attributedto the N-terminal segment of the αTP-subunit is associated withthe region between OH3 and OH4 (Fig. 2). Similar interactionsvia the αE- and αTP-subunits and the δ-subunit, the bacterialortholog of the OSCP, are present in the crystal structure of theF-ATPase from Paracoccus denitrificans (9).In P. angusta, and similarly in the bovine enzyme (SI Appendix,

Fig. S8), the C-terminal domain of the OSCP is predicted toconsist of a short β-strand, Oβ1 (residues 117–120), and twoα-helices, OH7 and OH8 (residues 130–141 and 176–194, re-spectively) with intervening β-strands, Oβ2 and Oβ3 (residues151–158 and 161–175). In the P. angusta structure, β1, OH7, andOH8 were resolved, but a region of density between OH7 andOH8 containing predicted Oβ2 and Oβ3 could not be interpreted(Fig. 2). However, the positions of selenomethionine residues,introduced biosynthetically into the truncated bovine peripheralstalk before its reassembly with bovine F1-ATPase to form bF1-tPS (18), coincide with the positions of equivalent residues inthe structure of the N-terminal domain of the P. angusta OSCP;the residues are Val-53, Leu-74, Leu-110, and Asn-111 (labeledO53, O74, O110, and O111; Fig. 2; SI Appendix, Fig. S8). Otherscoincide with the structure of OH8 in the C-terminal domain(residue Leu-190; O190 in Fig. 2 and SI Appendix, Fig. S8) and inthe b-subunit (Fig. 2; SI Appendix, Fig. S8) with b167 (Leu-167)and the adjacent pair b158/159 (Gln-158/Val-159). Therefore,O163, O166, and O177, which fall in regions of unmodeleddensity between OH7 and OH8, indicate where the corre-sponding residues (Leu-163 in the loop Oβ2-Oβ3, Leu-166 inOβ3, and Leu-177 probably just preceding OH8) lie in theP. angusta OSCP (Fig. 2). Residues 25–30 of the αDP-subunit areimmediately adjacent to where Oβ2 and Oβ3 are thought to be,and residues 11–22 of the αDP-subunit could interact with OH8,thus providing a third point of attachment of an α-subunit to theOSCP. The N-terminal region of the αDP-subunit probably alsointeracts with the b-subunit via bH2 and bH3 and with theh-subunit via hH1 (Fig. 2).The C-terminal domain of the OSCP is also bound to the

C-terminal region of the b-subunit. The membrane extrinsic partof the b-subunit from residues 70 to 204 is α-helical (SI Appendix,Fig. S9), and the α-helix in the bovine b-subunit is broken fromresidues 184 to 187 and is terminated by bH3 from residues 188 to207 (18). A similar structure is found in the P. angusta enzyme,where bH3 (residues 184–202) is associated with density in theregion between OH7 and OH8 (Fig. 2). Residues 165–179 at theC-terminal end of bH2 run alongside OH8, and the same region ofsubunit b is associated with the N-terminal α-helix of subunit h(hH1; residues 1–17). Subunit h is the fungal ortholog of bovine F6(SI Appendix, Fig. S10), and the interpretation of hH1 dependsupon the crystal structure of bF1-tPS (18). The remainder ofsubunit h is probably represented by unmodeled density that runsalong the outside of bH2 (SI Appendix, Fig. S11). The rest of themembrane extrinsic region of subunit b represented by bH2 is

about 160 Å long, and, in addition to subunit h, subunit d is alsoassociated closely with it. Subunit d is predicted to be folded intoseven α-helices, dH1–dH7 with extended segments (residues 49–57and 77–86) between dH3 and dH4 and dH5 and dH6, respectively(SI Appendix, Fig. S12). α-helices dH2–dH6 have similar structures tothose determined in the crystal structures of bF1-tPS and in theseparate bovine peripheral stalk (known as bPS). α-Helices dH4 anddH5 form a hairpin with an intervening turn from residues 63–67.Both bPS and bF1-tPS lacked the 36 C-terminal residues of subunitd corresponding to dH7 (132–138). The d-subunit of P. angusta has alonger C-terminal region and is predicted to have an extended dH7(residues 136–156) (SI Appendix, Fig. S12). Density for an α-helixadjacent to dH6 has been interpreted as dH7 running antiparallelto dH6, but the connection between them (residues 128–135) wasnot resolved. This region of the peripheral stalk, where α-helices in

A

B

O53

O74

O110

O111

O177

O190

b167

O163 and O166

b158 and b159

βTP αDP

αTP

b

OH7

OH8

h

αE

OH1OH2

αDP

αTP

OH4

αTP

αDP

βTP

βE

αE

βDP

h

Fig. 2. Attachment of the peripheral stalk to the crown of the F1-catalyticdomain of of the F-ATPase from P. angusta. The diagrams are based on state 1.(A and B) Views from the side and from above the N-terminal crown of theF1-domain in cartoon and surface representation, respectively. The OSCP issea-green, the three α-subunits are red, the three β-subunits are yellow, theb-subunit is pink, and subunit h is purple. In A, the N-terminal α-helical re-gions of the α-subunits are labeled αE, αDP, and αTP. In the OSCP, the positionsof α-helices OH1, OH2, OH4, OH7, and OH8 are indicated. The orangespheres represent the positions of selenium atoms in the structure of bovineF1-ATPase with the truncated peripheral stalk (bF1-tPS), and orange patchesindicate the positions of corresponding amino acids in the fungal OSCP. Theyare labeled according to the subunit where they reside, with the prefix O forOSCP and b for subunit b, followed by the residue number in the respectivesubunits. O53, O74, O110, O111, O190, b167, and b158–159 correspond toresidues in resolved regions of the structure of the P. angusta F-ATPase.O163, O166, and O177 indicate the positions of the selenomethionine resi-dues in bF1-tPS and indicate the positions of the equivalent P. angusta res-idues in unresolved regions of the current structure.

Vinothkumar et al. PNAS | November 8, 2016 | vol. 113 | no. 45 | 12711

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 3,

202

1

subunits b, d, and h lie parallel to each other, extends from near the“top” of the F1-domain, along the periphery of the F1-domain, to thelipid head group region of the membrane domain of the enzyme.However, the peripheral stalk is straighter in the P. angusta enzymethan in bF1-tPS, where it was bent toward the F1-domain by crystallattice contacts (18).

The Membrane Domain. Around residue 68, bH2 kinks and entersthe inner membrane of the mitochondrion, forming a trans-membrane α-helix from residues 51 to 68, in close agreementwith the secondary structure and hydrophobicity (SI Appendix,Fig. S9). A second transmembrane α-helix, bH1 is also predicted,and unconnected density lying parallel to bH2, has been attrib-uted to it, re-emerging from the matrix side of the membranearound residue 28. In the current model, α-helices bH1 and bH2interact with the a-subunit (described below), providing themembrane region of the stator with integrity. Based on theconservation of hydrophobic segments in their sequences, itis likely that the a-subunit itself has six transmembrane α-helices(aH1–aH6) as described before (7–9) (SI Appendix, Fig. S13). Inthe current structure, a bundle of four α-helices, each inclined at30° to the plane of the membrane, has been attributed to aH3–aH6; aH5 and aH6 consist of 36 and 37 residues, respectively,and they correspond to the exceptionally long hydrophobic seg-ments 5 and 6 in their sequences (SI Appendix, Figs. S13 andS14). α-Helix aH5 contains the absolutely conserved residue,Arg-179. Its positively charged side-chain is an essential featureof the transmembrane proton translocation mechanism (26) andis known to be close to another essential feature, the carboxyside-chain of Glu-59 in the c-subunit (27). The shorter α-helices,aH3 and aH4, have correspondingly shorter hydrophobic regionsin their sequences and are packed against aH5 and aH6. Thebiochemical data that support these and similar assignments in abacterial F-ATPase have been discussed elsewhere (9). Similarconclusions have been reached in the bovine enzyme (7). Densityattributed to aH2, joined by a loop to aH3, forms a relativelyshort α-helix that lies horizontally in the lipid head group region

of membrane on the matrix side, and aH1 is probably repre-sented by a vertical canonical α-helical segment (Ch3) packedagainst aH3 (Fig. 3).α-Helices aH5 and aH6 are intimately associated with the

c-ring that forms the membrane region of the enzyme’s rotor.The c-ring and the tightly associated central stalk together con-stitute the rotor. As in S. cerevisiae (22), the c-ring has 10 iden-tical c-subunits (Fig. 3), each with two transmembrane α-helicesjoined by a short loop on the matrix side of the inner mitochon-drial membrane. The N- and C-terminal α-helices form inner andouter concentric rings, as viewed in cross-section (Fig. 3). Fourc-subunits in the c-ring (I–IV in Fig. 3) are in contact with thesubunit a, mainly via inclined aH5 and aH6.The membrane domain of this particular preparation of the

monomeric F-ATPase from P. angusta contains three additionalsubunits, ATP8, f, and j, each with a single predicted trans-membrane α-helix (SI Appendix, Figs. S15 and S16) (20). Thesequence of ATP8 from P. angusta is related weakly to bovineATP8, and it is shorter (SI Appendix, Fig. S15). It is also relatedto the membrane domains of bacterial b-subunits (SI Appendix,Fig. S17). Its α-helical region has been ascribed to Ch2, brick redin Fig. 3A, with its C-terminal region protruding from the matrixof the inner mitochondrial membrane. The final feature resolvedin the membrane domain of the P. angusta enzyme is anothercanonical transmembrane α-helix (Ch1, pale yellow in Fig. 3A).This α-helix has been ascribed to subunit f rather than subunit jbecause subunit f has an extensive membrane extrinsic domain(SI Appendix, Fig. S16), which would account for some of theunascribed density in region I in SI Appendix, Fig. S11. Subunit jhas no such feature (SI Appendix, Fig. S16). Region I, the largestregion of unascribed density, also contains the N-terminal regionof the b-subunit and the membrane extrinsic region of ATP8.The membrane domain of the preparations of the P. angusta andbovine F-ATPases used in the cryo-EM studies differ in thecompositions of their supernumerary subunits. The former lackssubunits e, g, k, and l (20). Bovine mitochondria have no subunitsequivalent to k and l, but they contain orthologs of e and g, which

A

B D

aH3

aH6

aH5

aH4

In

OutaH2

aH1

R179

L176

L240L243

G212

L245K246

D247

Q233

E226

H188

N183

I IIIII

IVR129

S168

E165

C

Fig. 3. The membrane domain of the F-ATPase fromP. angusta. In A–D, the a-subunit is corn-flower blue.(A and B) Views in solid representation from the sideand below the membrane domain. The c10-ring isgray, the b-subunit (upper part not shown) is pink,and the pale yellow, brick-red, light cyan, and beigesegments are transmembrane α-helices, Ch1–Ch4assigned to subunit f, ATP8, aH1, and bH1, respec-tively. In the c-ring, I– IV indicate the four trans-membrane C-terminal α-helices in contact withsubunit a. (C and D) Views of the a-subunit in solidand cartoon representation viewed from outside andlooking out from the interface with the c-ring, re-spectively, with aH1 in pale cyan. Conserved polarresidues are yellow; the positions of human muta-tions associated with pathologies (SI Appendix, TableS1) are red. The pink sphere denotes the conservedArg-179 in aH5 that is essential for proton trans-location. The lower arrow indicates the inlet path-way for protons that transfer to Glu-59 in theC-terminal α–helix-II of the c-ring. They are carriedaround the ring by anticlockwise rotation, as viewedfrom above, until they arrive at Arg-179 where theyenter the exit pathway, denoted by the upper arrow.

12712 | www.pnas.org/cgi/doi/10.1073/pnas.1615902113 Vinothkumar et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 3,

202

1

were present in the preparation used in cryo-EM studies (7, 28).Therefore, a comparison of the P. angusta and bovine maps (SIAppendix, Fig. S18) indicates that the e- and g-subunits occupy anadditional region of density in the bovine map that probablycontributes to the formation of the dimeric enzyme in themitochondrial cristae.

DiscussionThe Mechanical Coupling Mechanism. The peripheral stalk of theF-ATPase is a central component that ensures the mechanicalcoupling of the transmembrane proton-motive force to the syn-thesis of ATP, and in this capacity it fulfills two main roles. First,it provides the enzyme’s stator with integrity by connecting theα3β3-catalytic domain to the a-subunit in the membrane domain,and those interactions have to be sufficiently strong to survivethe mechanical torque of the enzyme. Although the relevantinteractions are still not fully resolved, the current structuredemonstrates that the attachment to the catalytic domain ismuch more extensive and robust than had been thought, prob-ably involving both the N- and C-terminal domains of the OSCPand the N-terminal regions of all three α-subunits. The structurehas also provided evidence that, at the lower end of the pe-ripheral stalk (distal from the F1-domain), the hydrophobicN-terminal region of the b-subunit is, as predicted, folded into twotransmembrane α-helices, bH1 and part of bH2. It interacts withaH1 and aH2, and bH2 with aH5 and aH6, and most likely, inaddition, with the loop between aH3 and aH4. It has also con-firmed that the middle part of the peripheral stalk, consisting ofthe central α-helical pillar that is about 160 Å long and providedby bH2, is augmented by roughly parallel α-helices from subunitd and most likely from subunit h (as in the related bovine F6).Thus, it has the characteristics of a seemingly rigid and inflexiblestructure (29). The greatest uncertainty concerns part of theC-terminal domain of the OSCP and the region joining it to thecentral α-helical structure reaching down to the membrane,

where the density is currently not well resolved. It has beenproposed previously that this region could be flexible and pro-vide an elbow or joint between the N-terminal domain of theOSCP plus its attachments to the subunits αE and αTP and therigid α-helical region involving bH2 (18). Here bH3, which liesapproximately orthogonal to bH2, could be part of a pivot to-gether with OH7. The elbow, or joint, would allow the rigidα-helical pillar and attached membrane domain of the stator toadjust its position during a catalytic cycle, where the α3β3-domainis displaced from side-to-side by the turning of the asymmetricalupper region of the rotor. Therefore, it is of particular interestthat the region has been proposed to be the site where benzo-diazapine inhibitors bind (30, 31) and that Lys-139 of the OSCPbecomes acetylated and deacetylated in response to nutrient-and exercise-induced stress (32, 33), suggesting a regulatory rolefor the region.The second role of the peripheral stalk is to help keep the

a-subunit in contact with the rotating c-ring, possibly by exertinglateral pressure toward the central axis, thereby ensuring theintegrity of the transmembrane proton pathway. Subunit ATP8may contribute here via its C-terminal region. It is known thatthe longer C-terminal region of bovine ATP8 extends from themembrane into the peripheral stalk, where it interacts withsubunits b and d (34), thereby providing another brace in addi-tion to subunit b to hold subunit a against the rotating c-ring.

Conservation of the Proton Pathway. The structure of theP. angusta a-subunit and its interface with the c-ring are similarto those in the F-ATPases from P. denitrificans (9) and bovinemitochondria (7) (Fig. 3). The most striking feature in allthree structures is an inclined bundle of four α-helices made fromaH3–aH6 with aH5 and aH6 in contact with four adjacent ca-nonical transmembrane α-helices representing the C-terminalregions of four c-subunits in the c-ring. α-Helices aH3 and aH4are packed closely behind aH5 and aH6 (distal from the c-ring).As in the other structures (7, 9), aH5 contains the strictly con-served Arg-179 adjacent to Glu-59 (both P. angusta numbering)in C-terminal α-helix-III of the c-subunit. Both residues areknown from studies in Escherichia coli to be essential for protontranslocation (26, 27). Superimposition of the various structuresshows their overall similarity (SI Appendix, Fig. S19). However,their various structures are required to interact with c-rings fromc8 to c12, and the various ring sizes will require structural ad-justments in the a-subunit that are not readily apparent at thecurrent levels of resolution. Experimental evidence for the lo-cation of the proton path in the E. coli enzyme has been sum-marized previously (9). As noted before (9), the entry to theproton pathway is between aH5 and aH6, probably on the sidedistal from the c-ring, and the exit pathway also between aH5and aH6 is probably more proximal to the rotor. Both pathwayscontain conserved polar residues (Fig. 3D). The proposed pathwayis also consistent with human pathogenic mutations in subunit a(ATP6), which is encoded in the mitochondrial genome (35, 36)(SI Appendix, Table S1).

Perspectives.A comparison of the current structures of the bacterialand mitochondrial enzymes (Fig. 4) illustrates their known simi-larities in overall architecture and in the detailed structures of thecatalytic and proton translocating regions. Somewhat unexpectedly,the peripheral stalk regions of mitochondrial and bacterial enzymesare also similar, despite significant differences in subunit composi-tion (17, 18, 37–41) and lack of similarity in sequence of theirconstituent subunits (excepting the orthologous mitochondrialOSCP and bacterial δ-subunits). The relationship between theATP8 subunit and the membrane domains of the bacterial b-sub-units (SI Appendix, Fig. S17) adds to this structural similarity. Thus,peripheral stalks from bacteria, chloroplasts, and eukaryotes havesimilar designs, and presumably similar physical properties, to allow

A B C

Fig. 4. Comparison of the current structures of the F-ATPases. The struc-tures of the three complexes are viewed from the side toward the peripheralstalk with the F1-catalytic domain above and the membrane domain be-neath. Enzymes (A and B) from P. angusta and bovine mitochondria (7) and(C) from P. denitrificans (9). The α-, β-, and γ-subunits are red, yellow, and royalblue, respectively. The c-rings (made of 10, 8, and 12 subunits in A–C, re-spectively) are gray, and the adjacent a-subunits are corn-flower blue. InA and B, the OSCP subunits (Top), and in C, the orthologous δ-subunit, aresea-green; the δ-subunits (A and B) and orthologous e-subunit in C are green.In A and B only, the e-subunit (next to the green δ-subunit) is magenta. In Aand B, in addition to the OSCP, the peripheral stalks contain the b-subunits(pink), the d-subunits (orange), and the orthologous h and F6-subunits(purple), respectively. In C, in addition to the δ-subunit, the peripheral stalkcontains a b-subunit (pink) and a b′-subunit (orange). In A, the pale yellowand brick-red α-helices packed against the c-rings have been assigned ten-tatively to the f and ATP8 subunits, respectively.

Vinothkumar et al. PNAS | November 8, 2016 | vol. 113 | no. 45 | 12713

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 3,

202

1

them to perform their roles in ensuring the maintenance of an intactproton pathway in the interface between rotor and stator and inkeeping the stator together. The current descriptions of the pathwayitself are rudimentary, and a full understanding will require atomicresolution structures where positions of amino acid side-chains andparticipating water molecules are defined. Also to be taken intoconsideration is the role of lipids in the mechanism of the enzyme.Cardiolipin bearing two negative charges is an essential componentof active F-ATPases in bacteria and mitochondria (42). It appearsto be attracted to the c-ring selectively over phospholipids, where itbinds transiently and repeatedly around basic residues in the head-group regions of both leaflets of the bilayer, suggesting a possiblerole in proton translocation (42).As this paper neared completion, the structure of the dimeric

F-ATPase from the fungus Yarrowia lipolytica determined by cryo-EM was published (43). There is excellent agreement betweenmany of the features in the two enzymes. The current paper con-tains a more precise and accurate description of the attachment

of the peripheral stalk to the catalytic domain, and the dimericstructure confirms the presence of the e- and g-subunits in thedimer interface and gives undiscovered details of how theyare folded.

Materials and MethodsThe F-ATPase from P. angusta was purified as described before (20). Imagesof single monomeric complexes were recorded with a Titan Krios electronmicroscope (FEI) with a Falcon II CMOS (complementary metal oxide semi-conductor) direct electron detector using the EPU automated data acquisitionsoftware. For full details of these processes and the structure determination andanalysis, see SI Appendix, Materials and Methods.

ACKNOWLEDGMENTS. We thank Drs. A. Leslie and R. Henderson for their com-ments; S. Chen and C. Savva for help with electron microscopes; and J. Grimmettand T. Darling for help with computing. This work was supported by the MedicalResearch Council of the United Kingdom by Grant MC_U1065663150 and by Pro-gramme Grant MR/M009858/1, (both to J.E.W.) and by Grant MC_U105184322(to K.R.V.).

1. Watanabe R, Noji H (2013) Chemomechanical coupling mechanism of F1-ATPase:Catalysis and torque generation. FEBS Lett 587(8):1030–1035.

2. Walker JE (2013) The ATP synthase: The understood, the uncertain and the unknown.Biochem Soc Trans 41(1):1–16.

3. Robinson GC, et al. (2013) The structure of F₁-ATPase from Saccharomyces cerevisiaeinhibited by its regulatory protein IF₁. Open Biol 3(2):120164.

4. Bason JV, Montgomery MG, Leslie AGW, Walker JE (2014) Pathway of binding of theintrinsically disordered mitochondrial inhibitor protein to F1-ATPase. Proc Natl AcadSci USA 111(31):11305–11310.

5. Bason JV, Montgomery MG, Leslie AGW, Walker JE (2015) How release of phosphatefrom mammalian F1-ATPase generates a rotary substep. Proc Natl Acad Sci USA112(19):6009–6014.

6. Baker LA, Watt IN, Runswick MJ, Walker JE, Rubinstein JL (2012) Arrangement ofsubunits in intact mammalian mitochondrial ATP synthase determined by cryo-EM.Proc Natl Acad Sci USA 109(29):11675–11680.

7. Zhou A, et al. (2015) Structure and conformational states of the bovine mitochondrialATP synthase by cryo-EM. eLife 4:e10180.

8. Allegretti M, et al. (2015) Horizontal membrane-intrinsic α-helices in the statora-subunit of an F-type ATP synthase. Nature 521(7551):237–240.

9. Morales-Rios E, Montgomery MG, Leslie AGW, Walker JE (2015) Structure of ATPsynthase from Paracoccus denitrificans determined by X-ray crystallography at 4.0 Åresolution. Proc Natl Acad Sci USA 112(43):13231–13236.

10. Arnold I, Pfeiffer K, Neupert W, Stuart RA, Schägger H (1998) Yeast mitochondrialF1Fo-ATP synthase exists as a dimer: Identification of three dimer-specific subunits.EMBO J 17(24):7170–7178.

11. Paumard P, et al. (2002) Two ATP synthases can be linked through subunits in theinner mitochondrial membrane of Saccharomyces cerevisiae. Biochemistry 41(33):10390–10396.

12. Arselin G, et al. (2003) The GxxxG motif of the transmembrane domain of subunit e isinvolved in the dimerization/oligomerization of the yeast ATP synthase complex inthe mitochondrial membrane. Eur J Biochem 270(8):1875–1884.

13. Fronzes R, Weimann T, Vaillier J, Velours J, Brèthes D (2006) The peripheral stalkparticipates in the yeast ATP synthase dimerization independently of e and g sub-units. Biochemistry 45(21):6715–6723.

14. Dudkina NV, Heinemeyer J, Keegstra W, Boekema EJ, Braun HP (2005) Structure ofdimeric ATP synthase from mitochondria: An angular association of monomers in-duces the strong curvature of the inner membrane. FEBS Lett 579(25):5769–5772.

15. Strauss M, Hofhaus G, Schröder RR, Kühlbrandt W (2008) Dimer ribbons of ATP syn-thase shape the inner mitochondrial membrane. EMBO J 27(7):1154–1160.

16. Collinson IR, Skehel JM, Fearnley IM, Runswick MJ, Walker JE (1996) The F1Fo-ATPasecomplex from bovine heart mitochondria: The molar ratio of the subunits in the stalkregion linking the F1 and Fo domains. Biochemistry 35(38):12640–12646.

17. Dickson VK, Silvester JA, Fearnley IM, Leslie AGW, Walker JE (2006) On the structureof the stator of the mitochondrial ATP synthase. EMBO J 25(12):2911–2918.

18. Rees DM, Leslie AGW, Walker JE (2009) The structure of the membrane extrinsic re-gion of bovine ATP synthase. Proc Natl Acad Sci USA 106(51):21597–21601.

19. Ueno H, Suzuki T, Kinosita K, Jr, Yoshida M (2005) ATP-driven stepwise rotation ofFoF1-ATP synthase. Proc Natl Acad Sci USA 102(5):1333–1338.

20. Liu S, et al. (2015) The purification and characterization of ATP synthase complexesfrom the mitochondria of four fungal species. Biochem J 468(1):167–175.

21. Gledhill JR, Montgomery MG, Leslie AGW, Walker JE (2007) How the regulatoryprotein, IF1, inhibits F1-ATPase from bovine mitochondria. Proc Natl Acad Sci USA104(40):15671–15676.

22. Stock D, Leslie AGW, Walker JE (1999) Molecular architecture of the rotary motor inATP synthase. Science 286(5445):1700–1705.

23. Abrahams JP, Leslie AGW, Lutter R, Walker JE (1994) Structure at 2.8 Å resolution ofF1-ATPase from bovine heart mitochondria. Nature 370(6491):621–628.

24. Kagawa R, Montgomery MG, Braig K, Leslie AGW, Walker JE (2004) The structureof bovine F1-ATPase inhibited by ADP and beryllium fluoride. EMBO J 23(14):2734–2744.

25. Bowler MW, Montgomery MG, Leslie AGW, Walker JE (2007) Ground state structureof F1-ATPase from bovine heart mitochondria at 1.9 Å resolution. J Biol Chem 282(19):14238–14242.

26. Lightowlers RN, Howitt SM, Hatch L, Gibson F, Cox GB (1987) The proton pore in theEscherichia coli FoF1-ATPase: A requirement for arginine at position 210 of thea-subunit. Biochim Biophys Acta 894(3):399–406.

27. Hoppe J, Sebald W (1980) Amino acid sequence of the proteolipid subunit of theproton-translocating ATPase complex from the thermophilic bacterium PS-3. Eur JBiochem 107(1):57–65.

28. Runswick MJ, et al. (2013) The affinity purification and characterization of ATP syn-thase complexes from mitochondria. Open Biol 3(2):120160.

29. Sielaff H, et al. (2008) Domain compliance and elastic power transmission in rotaryFoF1-ATPase. Proc Natl Acad Sci USA 105(46):17760–17765.

30. Johnson KM, et al. (2005) Identification and validation of the mitochondrial F1Fo-ATPaseas the molecular target of the immunomodulatory benzodiazepine Bz-423. Chem Biol12(4):485–496.

31. Cleary J, Johnson KM, Opipari AW, Jr, Glick GD (2007) Inhibition of the mitochondrialF1Fo-ATPase by ligands of the peripheral benzodiazepine receptor. Bioorg Med ChemLett 17(6):1667–1670.

32. Wu YT, Lee HC, Liao CC, Wei YH (2013) Regulation of mitochondrial FoF1 ATPaseactivity by Sirt3-catalyzed deacetylation and its deficiency in human cells harboring4977bp deletion of mitochondrial DNA. Biochim Biophys Acta 1832(1):216–227.

33. Vassilopoulos A, et al. (2014) SIRT3 deacetylates ATP synthase F1 complex proteins inresponse to nutrient- and exercise-induced stress. Antioxid Redox Signal 21(4):551–564.

34. Lee J, et al. (2015) Organisation of subunits in the membrane domain of the bovineF-ATPase revealed by covalent cross-linking. J Biol Chem 290(21):13308–13320.

35. Kucharczyk R, et al. (2009) Mitochondrial ATP synthase disorders: Molecular mecha-nisms and the quest for curative therapeutic approaches. Biochim Biophys Acta1793(1):186–199.

36. Xu T, Pagadala V, Mueller DM (2015) Understanding structure, function, and muta-tions in the mitochondrial ATP synthase. Microb Cell 2(4):105–125.

37. Walker JE, Saraste M, Gay NJ (1984) The unc operon. Nucleotide sequence, regulationand structure of ATP-synthase. Biochim Biophys Acta 768(2):164–200.

38. Dunn SD (1992) The polar domain of the b subunit of Escherichia coli F1Fo-ATPaseforms an elongated dimer that interacts with the F1 sector. J Biol Chem 267(11):7630–7636.

39. Collinson IR, et al. (1994) ATP synthase from bovine heart mitochondria. In vitro as-sembly of a stalk complex in the presence of F1-ATPase and in its absence. J Mol Biol242(4):408–421.

40. Karrasch S, Walker JE (1999) Novel features in the structure of bovine ATP synthase.J Mol Biol 290(2):379–384.

41. Dunn SD, McLachlin DT, Revington M (2000) The second stalk of Escherichia coli ATPsynthase. Biochim Biophys Acta 1458(2-3):356–363.

42. Duncan AL, Robinson AJ, Walker JE (2016) Cardiolipin binds selectively but transientlyto conserved lysine residues in the rotor of metazoan ATP synthases. Proc Natl AcadSci USA 113(31):8687–8692.

43. Hahn A, et al. (2016) Structure of a complete ATP synthase dimer reveals themolecular basis of inner mitochondrial membrane morphology. Mol Cell 63(3):445–456.

12714 | www.pnas.org/cgi/doi/10.1073/pnas.1615902113 Vinothkumar et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 3,

202

1